The bandwidth requirements for machine vision, as in almost every other electronic discipline, are increasing. Image sensors now output ever larger amounts of data, which has to be transferred from camera to PC. Teledyne Dalsa’s top-end Piranha3 16k line scan camera has a throughput of 1.146 Gigapixels per second and other sensors are available outputting in the Gigapixel per second region. This is outside the range of Camera Link Full (6.1Gb/s), the connectivity standard traditionally considered for high-speed industrial vision applications.
Two new machine vision standards are coming through, addressing some of the bandwidth issues generated by the latest breed of image sensor technology. CoaXPress version 1.0, hosted by the Japan Industrial Imaging Association (JIIA), has recently been released at the Automate show in Chicago as a global standard supported by the Automated Imaging Association (AIA) and the European Machine Vision Association (EMVA), and Camera Link HS, still under development, is expected to be released later in the year. In addition, GigE Vision version 2.0, which also has a release date later this year, will have support for 10 Gigabit Ethernet, effectively increasing its bandwidth capabilities tenfold to 10Gb/s, and link aggregation, which further increases the bandwidth by coupling a number of cables together (40Gb/s with four cables).
Of course, it’s not all about bandwidth, so how will users decide on a standard that meets their needs? Apart from bandwidth, cable length and cost are the other big differentiators. Then there are aspects like signal latency and jitter, and real-time triggering to consider.
One of the big benefits of CoaXPress identified by Mark Williamson, director of corporate market development at Stemmer Imaging, is that the standard uses single-wire coaxial cable for most applications, with power, data transfer and timing down to microsecond accuracies, all on a single cable. ‘One of the most interesting aspects is that you can replace old analogue cameras with high-resolution digital cameras using the same cable network,’ he says. ‘This is particularly apt for traffic applications such as the London congestion charge system, which was installed five or so years ago all using coax cabling with standard video cameras. With CoaXPress, the camera network could be retrofitted with high-resolution digital cameras enabling automatic number plate recognition (ANPR) and other image analysis.’
CoaXPress was initially developed by a consortium of companies made up of Adimec, EqcoLogic, Active Silicon, Components Express, Aval Data, and NED. It won the Vision Award at the Vision show in Stuttgart in 2009 and now more than 75 companies worldwide are members of the CoaXPress working groups. Jochem Herrmann, chief scientist at Adimec and chair of the CoaXPress consortium, sees a wide range of applications that would benefit from the standard, from companies wanting simpler cabling that could be used with a slip ring, for instance (CoaXPress is a single cable comprised of two wires, which lends itself to slip rings for fully-rotating devices, while less flexible cabling such as that for Camera Link doesn’t), to those that would like to go faster than GigE, but not be limited by the 10m cable lengths of Camera Link. ‘Coaxial cable has a huge capacity if you’re able to drive it and receive it well,’ Herrmann states. CoaXPress is able to transfer data at a rate of 6.25Gb/s on a single coax cable and can be scaled to higher speeds by adding more cables. ‘Ten years ago, you could only dream of this – the state-of-the-art solutions were 20-30 times slower,’ Herrmann says.
CoaXPress uses an automatic equaliser from EqcoLogic that monitors and automatically corrects for the attenuation and frequency. Therefore, at the output of the equaliser the original waveform is automatically restored. ‘FPGAs nowadays have fast serialisers and deserialisers allowing rapid transfer of digital information over one cable,’ notes Herrmann. ‘The latest driver and equaliser technologies make it possible to transport these high data rates over long cables.’
In terms of overall cost, Herrmann feels that, at the moment, there is no difference between CoaxExpress and Camera Link. ‘Camera Link is simple and cheap to implement in the camera, but it uses expensive cables. CoaXPress is somewhat more expensive to implement in the camera – the chips are more expensive – but the coaxial cable is very cheap,’ he says, although adding that he expects CoaXPress solutions will become cheaper than Camera Link in the near future as the volume of CoaXPress products grow. ‘CoaXPress is not designed to be low cost, but the customer will benefit from the simpler and cheaper cabling, ease of installation, and longer length. The application will dictate, to a large extent, which interface is used,’ Herrmann states.
Need for speed
While CoaXPress caters for reasonably high-bandwidth applications, matching Camera Link Full data transfer capacity on one coax cable with the ability to scale up by adding more cables, Williamson of Stemmer Imaging feels that for very high-end solutions, Camera Link HS is the connectivity standard of choice: ‘Some of Teledyne Dalsa’s top-end cameras go beyond the speed of CoaXPress. When you really start to go to the top end, CoaXPress becomes less desirable and Camera Link HS is more suitable.’
Comparison of machine vision standards
Camera Link HS sprung from HS Link, a precursor to the standard pioneered by Dalsa and developed to operate some of its top-end cameras. The standard is being run by the Automated Imaging Association (AIA), which formed the Camera Link HS committee in April 2010.
‘Dalsa HS Link was proof of concept and we’ve been selling products based on this for a year and a half,’ comments Michael Miethig, technical manager at Teledyne Dalsa and chair of the Camera Link HS committee.
‘Our customers are very happy with HS Link. They use 10-20 of these [Gigapixel per second] cameras in an inspection system, so if they can reduce their cabling they’re happy.’ One CX-4 cable achieves data transfer rates of 2.1GB/s.
According to Miethig the protocol will scale to single lanes as well. ‘We’ve approved a fibre optic SFP connection, which is very cost-effective. The protocol is designed to scale from one lane, about 300MB/s effective bandwidth using 3.125Gb/s 8b/10b encoding, up to as many lanes as you need.’
Miethig says that the features in Camera Link HS are based on those in Camera Link, i.e. low latency and real-time triggering. ‘Out of all the serial protocols, Camera Link HS has the lowest latency and lowest jitter [in the nanosecond range],’ he states. In comparison, GigE Vision with IEEE-1588 cabling is in the tens or hundreds of microseconds latencies.
Camera Link HS is currently entering into a pre-release phase – Mikrotron and Silicon Software will release products that use the unofficial headers soon, according to Miethig. He comments: ‘I’m hoping that by the Vision show in Stuttgart in November 2011 we will have debugged all the various aspects and we’ll release version 1.0 of the standard.’
Established solutions
GigE Vision also proposes to go to higher bandwidths with the planned release of version 2.0 of the standard supporting 10 Gigabit Ethernet and link aggregation. Version 2.0 will also include support for real-time protocol (RTP), which allows microsecond synchronisation (±0.5μs timing accuracy) between cameras across Ethernet.
‘One of the problems with older versions of GigE Vision was that the communication wasn’t deterministic and therefore the cameras had to be triggered directly via trigger cables,’ comments Williamson. ‘This was one of the limitations of previous versions of GigE Vision. You could send the data, but there was no deterministic timing control. With the RTP, cameras will be able to synchronise their internal time bases.’
The advantage of GigE Vision is that the user doesn’t need a dedicated acquisition device, whereas Camera Link, Camera Link HS, and CoaXPress are frame grabber-based architectures.
Refinements have also been made to the existing Camera Link standard – version 2.0 was released in March 2011 – which, according to Marc Damhaut, senior VP product management at frame grabber manufacturer Euresys, allows smaller and potentially cheaper frame grabbers to be produced.
The company’s latest cards, the Grablink Full and the Grablink DualBase, contain two Mini Camera Link connectors and the typical I/O connector on the bracket. ‘This is due to the updated standard, which specifies smaller connectors than the original version,’ Damhaut says.
However, one of the problems with Camera Link that Damhaut identifies is that it only standardises the connector, the cable and the pin-out of the cable. ‘I would prefer a standard that better defines the protocols involved in the exchange of data with the camera, which isn’t included in Camera Link, but is included in GigE Vision and FireWire,’ he says.
USB 3.0 products are also now being released (USB 3.0 will reach 5Gb/s). Williamson feels that USB 3.0 is not a true machine vision interface, but that it does have benefits over the other standards in that it is plug-and-play and included on every PC. However, USB cameras have relatively short cable length, with USB 3.0 limited to 3m. ‘It’s great for connecting a camera to a microscope, but for industrial vision, USB 3.0 is currently limited,’ Williamson says.
What the future holds
So what will the machine vision connectivity landscape look like in the future? Most commentators feel that there is a place for both CoaXPress and Camera Link HS in the market. ‘CoaXPress and Camera Link HS are going to be there to complement GigE Vision,’ comments Pierantonio Boriero, product line manager at Matrox Imaging. ‘They’re going to address a niche where the amount of data that needs to be transmitted and the real-time aspect exceeds the capabilities of GigE Vision.’ Matrox Imaging is part of the AIA working groups for both CoaXPress and Camera Link HS and it is also actively designing products for CoaXPress. As with most things, it will ultimately come down to application. ‘Every standard has some unique features that might be most advantageous in certain applications,’ states Miethig of Teledyne Dalsa.
Teledyne Dalsa’s top-end Piranha3 16k line scan camera outputs in the Gigapixel per second region, a driver for the development of Camera Link HS
Williamson says Camera Link HS is also the only standard that will go to the very top end and Miethig says the standard is designed to migrate to 10Gb/s if necessary: ‘It’s currently at 3.125Gb/s, but PCO is developing a 10Gb/s version, so we know we’re going faster. We’re not sure if we’ll go slower.’
Boriero, however, says that for many low- to medium-bandwidth applications, depending on the customer’s real-time I/O requirements, GigE Vision would still be suitable.
Williamson believes CoaXPress will be successful because of the retrofit market in areas like medical and defence.
One potential stumbling block with CoaXPress is that the standard was initially based on single-source transceiver chips from EqcoLogic. François Bertrand, vice president, sales and marketing at Matrox Imaging, says: ‘Despite its interesting technical capabilities, CoaXPress, like other technologies designed for machine vision and industrial applications, has a latent issue with the sourcing of the receiver/transceiver part from a single and relatively modest player in the semiconductor world. The JIIA is aware of this limitation and is working to address the matter both technically and commercially to ensure wide acceptance of this emerging standard.’
Herrmann of Adimec points out that all new technologies start as single source. He goes on to say that to minimise risks, JIIA has arranged an Escrow agreement with EqcoLogic, which guarantees supply if anything happens to the company. In addition to that, CoaXPress solutions have been developed using chips of National Semiconductor and Gennum, making the standard much more robust.
A potential disadvantage with Camera Link HS, says Williamson, is that there is pressure not to adopt a software standard transport interface, as is the case with CoaXPress, which has fully supported the GenICam TL standard. What this would mean is that each hardware interface would require its own proprietary API, as with Camera Link. ‘While Camera Link HS looks to outperform CoaXPress in terms of speed, this omission could make Camera Link HS less desirable unless speed dictates its use,’ he states. ‘I think this is a missed opportunity in getting to one software interfacing standard across all machine vision interfaces.’
There are also other potential new standards on the horizon for the machine vision industry. A new interface has recently been released by Intel in collaboration with Apple called Thunderbolt (formerly known as Light Peak). It uses a USB-type connector to transfer data at 10Gb/s. ‘This will be interesting in the future for machine vision technology, maybe in two to three years’ time,’ Williamson states, although he adds that, as with USB, it’s not an industrial machine vision standard. We will wait to see though the progress of CoaXPress and Camera Link HS as camera manufacturers begin to develop products for them.